Identification of Stromal Factors Important for BCC Tumorigenesis. To identify factors produced by tumor-associated stromal cellsthat contribute to the initiation or maintenance of BCC, wecultured stromal cells from fresh samples of human BCC or nontumorskin. The cells adhered readily to untreated plastic platesand were spindle-shaped and elongated. We used the SignificanceAnalysis of Microarrays algorithm to identify genes differentiallyexpressed between tumor- and nontumor-associated stromal cells(Fig. 1A; ref. 29). Fourteen genes were identified at an estimatedfalse discovery rate of 5%, with 13 genes expressed at higherlevels and one gene expressed at a lower level in the tumor-associatedcells (Fig. 1B and Table 1, which is published as supporting informationon the PNAS web site). Two of the 13 genes more highly expressedin BCC-derived stromal cells, GREM1 (GREMLIN 1) and FST (FOLLISTATIN),both encode antagonists of the BMP pathway.
GREMLIN 1 Expression Is Elevated in BCC, and BMPs Are Highly Expressed by BCC Tumors. We analyzed expression of GREMLIN 1 in vivo in human tissueby quantitative RT-PCR analysis of independent samples of wholetissue from eight matched BCC and adjacent nontumor skin samples.GREMLIN 1 transcripts were, indeed, expressed at higher levelsin BCC tissue than in adjacent nontumor tissue from the samepatient (Fig. 1C). We then performed in situ hybridization (ISH)in 15 paraffin-embedded BCC tissue samples and found detectableGREMLIN 1 mRNA expression in 12 of 15 samples (80%). Expressionwas localized predominantly to stromal cells in the tumor, andimmunohistochemistry (IHC) localized gremlin 1 protein to thestroma surrounding the tumor cell nests (Fig. 2B and D and Fig.6, which is published as supporting information on the PNASweb site). In contrast, no expression of GREMLIN 1 RNA or proteinwas detected in normal skin (Fig. 2 A and C). Thirty-nine sectionsof normal skin from multiple anatomical sites, including arm(dorsal, ventral, posterior, and anterior), hand (dorsal andventral), digits (posterior), palm, foot (dorsal and plantar),and leg (anterior, posterior, dorsal, and midline), were allnegative for GREMLIN 1 RNA, with only two exceptions: a fewstromal cells surrounding a neuromuscular junction in one sectionof skin from below the knee, and a small number of stromal cellsdeep in the dermis of the foot dorsum (data not shown). Theseresults indicate that GREMLIN 1 RNA expression is below levelsof detection or absent in the vast majority of normal humanskin sites.
An implicit aspect of our hypothesis is that there exists asource of BMP in BCC tumors that needs to be antagonized topromote proliferation of tumor cells. We found that BMP 2 and4 are, indeed, expressed in BCC tumor nests (Fig. 2 E and F).BMP antibody staining localized mostly to tumor cells, withmacrophages occasionally demonstrating positive staining.
To better characterize the stromal cell population that expressedGREMLIN 1 in BCC tumors, we analyzed adjacent serial sectionsof tumor by ISH for GREMLIN 1 and IHC for various cell lineagemarkers: vimentin (characteristic of mesenchymal cells), CD45(hematopoietic lineage), CD31 (endothelial cells), desmin (smoothmuscle cells), cytokeratins (epithelial cells), and glial fibrillaryacid protein (astrocytes and Schwann cells). GREMLIN 1-expressingcells were also strongly positive for vimentin, mostly or entirelynegative for CD45 and desmin, and completely negative for CD31,keratins, and glial acid fibrillary protein (GFAP) (Fig. 2 G–N).
A Functional Response to gremlin 1 in Cultured Human Skin Epithelial Cells. We reasoned that if the functional role of gremlin 1 in maintaininga tumor cell niche was analogous to its role in the normal skinprogenitor cell niche, gremlin 1 might be capable of inhibitingdifferentiation and promoting expansion of keratinocytes. Todirectly examine the effects of gremlin 1 on BCC tumor cells,cells isolated from fresh BCC tumors were cultured in the presenceof recombinant human BMP 4, recombinant mouse gremlin 1, orboth, and allowed to expand for 7 days. The resulting cell populationswere compared by using quantitative RT-PCR to characterize theirdifferentiation state (Fig. 3A). Compared with untreated controls,cells maintained in BMP 4 exhibited elevated mRNA levels ofSPRR1A, SPRR1B, SPRR3, and SPRR4, established markers of differentiatedkeratinocytes. Gremlin 1 strongly attenuated this effect. Gremlin1 protein alone, in the absence of exogenously added BMP 4,had little effect on SPRR expression. [Note that basal mediacontains no detectable BMP (data not shown).]
Gremlin 1 also antagonized BMP-mediated repression of cell proliferation.Primary BCC keratinocytes were cultured and cell growth assessedin the presence of varying concentrations of recombinant humanBMP 4 and recombinant mouse gremlin 1 (Fig. 3B). The doublingtime of these cells in culture with no added BMP or gremlin1 was 3.1 (+/– 0.1) days. Addition of gremlin 1 in theabsence of added BMP 4 did not significantly affect growth rate,even at the highest concentration of gremlin (2.105 µg/ml).In the absence of gremlin 1, doubling time increased steadilywith increasing BMP 4 concentration, reaching a maximum of 7.4(+/– 0.1) days, 2.4 times the baseline doubling rate.At the highest level of BMP 4, increasing the concentrationof gremlin 1 protein steadily lowered the doubling time backto baseline. These results indicated that BMP 4 inhibits theexpansion of BCC cell populations in culture, and that gremlin1 attenuates this inhibition.
GREMLIN 1 Is Expressed by Stromal Cells in Diverse Human Carcinomas. GREMLIN 1 is highly expressed in the fibroblasts of most BCCsand undetectable in most normal skin sites. Evidence that BMPsregulate stem cell expansion in many tissues (skin, intestine,and blood) raised the possibility that expression of gremlin1 may be an important feature of the tumor microenvironmentin other cancers (12, 24, 25). We therefore examined GREMLIN1 RNA expression in a total of 774 tumors, including melanomaand carcinomas of the liver, testis, ovary, uterus, kidney,thyroid, prostate, head and neck, bladder, breast, lung, colon,pancreas, and esophagus (n = 11–260 samples of each) byISH to tissue microarrays. GREMLIN 1 was expressed by stromalcells in at least 50% of samples in carcinoma of the bladder,breast, lung, colon, pancreas, and esophagus, and in at least25% of prostate and head and neck cancers (Fig. 4). Expressionof GREMLIN 1 was exclusively localized to the stromal cells,with the exception of some breast and prostate samples, whichshowed limited expression in the tumor cells themselves.
We also examined large sections of breast, pancreas, lung, andintestine, both tumor and nontumor. GREMLIN 1 expression wasundetectable in normal and benign breast tissue. In a seriesof 165 samples of pancreas, including normal tissue and benignand malignant lesions, we detected GREMLIN 1 RNA in only 5%(2/37) of normal samples, compared with 71.5% of pancreatictumors (68/95) (Fig. 7, which is published as supporting informationon the PNAS web site). GREMLIN 1 expression was also detectedin 45% (15/33) of benign pancreatic disease samples, includingpancreatitis, benign neuroendocrine tumors, and benign adenomas.In normal lung tissue, we observed GREMLIN 1 RNA in only a fewsmooth muscle cells. In large sections of both adenocarcinomaof the lung and adjacent normal lung tissue, there was no detectableGREMLIN 1 mRNA in the normal lung, whereas the tumor stromaand not the tumor cells themselves showed expression of GREMLIN1 mRNA (Fig. 8, which is published as supporting informationon the PNAS web site). In normal intestine, no GREMLIN 1 expressionwas observed except in the lamina propria, in what appear bymorphology to be smooth muscle cells (data not shown).